Handheld manipulation gripper

• I built the gripper as a parametric, history-based model in Onshape: a 16-part assembly with mates driving the jaw linkage, from sketch-based extrusions and boolean cuts. 3 of those parts are mine rather than the reference design's, an Arduino Nano holder, an MPU6050 IMU tray and a webcam cover, plus their lids.
• We researched a tetrahedral form and built our own grippers off it. They grip 43% more than an average gripper.
• V2 came down from 636 g to 200 g, a 3.2× reduction, through topology optimization and generative design on the structural members.
• We designed our own PCB around a Raspberry Pi Zero for the sensing stack.
• It matches the end effector on the robot it trains, so recorded data transfers without a retargeting step. Workers wear it through an ordinary shift instead of staging demonstrations.
• Design for manufacturing: I audited all 16 parts for overhangs past 45° in their print orientations. 7 came back clean, reorientation cleared 4, and 45° gussets cleared 1 more. Standing the handle upright took ~70 cm² of internal ceiling down to ~3 cm² of sub-50° slope, and flipping the cube avoided a 65 × 90 mm supported ceiling outright.
• The remaining 3 carry breakaway scaffolding I designed into the STL: 7 comb rows of 2.8 mm tapered posts necking to 1.2 mm, 0.8 mm tips, 0.5 mm release gap. Slicer rule for all 16: supports off.
• Tolerance stack-up: the AprilTag fiducials seat in 20.00 × 20.00 mm recesses, verified against a printed scale-check bar on every machine.
• Runs a 0.6 mm nozzle at 0.3 mm layers, 100% infill and 1.8 mm walls on the load-bearing grip, 2× MGN9C 150 mm linear rails on the jaws, and 26 M3 fasteners across 4 lengths.
• Built for mass UMI data collection from skilled laborers in India, to sell to robotics foundation model labs.

Code for classes still running is kept private under Berkeley's academic-honesty policy. Happy to walk through it in person. ← Back to all projects